Identifying consistent allele frequency differences in studies of stratified populations

Area of Science:

• Population genomics
• Evolutionary biology
• Bioinformatics

Background:

• Pooled sequencing is increasingly used in population genomics.
• Accurate quantification of allele frequency differences between populations is essential for detecting genomic regions under selection.
• Existing statistical tests often violate underlying assumptions, leading to unreliable results.

Purpose of the Study:

• To evaluate the performance of common statistical tests for allele frequency differences.
• To introduce and assess alternative methods, such as quasibinomial generalized linear models (GLMs), for analyzing allele frequency data.
• To address issues like high false positive rates and the impact of sequencing coverage variation.

Main Methods:

• Simulation of population genetic models under neutral evolution.
• Assessment of the performance of various statistical tests, including popular methods and GLMs with quasibinomial error structure.
• Re-analysis of a published dataset using the proposed methods.

Main Results:

• Common statistical tests exhibit poor performance and high false positive rates in simulations.
• GLMs with quasibinomial error structure demonstrate improved inference by avoiding confounding of heterogeneity and main effects.
• Adjusting allele frequencies by re-scaling to a common value or effective sample size reduces false positives caused by sequencing coverage variation.

Conclusions:

• Quasibinomial GLMs offer a more robust approach for analyzing allele frequency differences in population genomics.
• These models provide better control over false positives and can be readily extended to more complex experimental designs.
• Accurate methods are vital for identifying loci associated with phenotypic variation and adaptation.